genimg

genimg — Image Generation

Text-to-image and image-to-image via any OpenAI-compatible image API.

Usage

/genimg <prompt>                     # text-to-image
/genimg <image_path> <prompt>        # image-to-image
/genimg init                         # interactive setup wizard

Setup

If ~/.genimg/config.json doesn't exist, the script will print a friendly error with the expected format and suggest running /genimg init.

/genimg init — Interactive Setup

When the user runs /genimg init, walk them through creating or updating ~/.genimg/config.json:

1. Ask for provider name (e.g. "volcengine", "openai", "together")
2. Ask for base URL (the OpenAI-compatible API endpoint, e.g. https://ark.cn-beijing.volces.com/api/v3)
3. Ask for API key
4. Ask for model name (e.g. doubao-seedream-3-0-t2i-250415, dall-e-3)
5. Ask for default size (e.g. 2048x2048, 1024x1024)
6. Write config to ~/.genimg/config.json, setting this as default_provider

If a config already exists, show current providers and offer to add a new one or update an existing one.

Config

Provider settings live in ~/.genimg/config.json. Structure:

{
  "providers": {
    "<name>": {
      "base_url": "...",
      "api_key": "...",
      "model": "...",
      "default_size": "2048x2048"
    }
  },
  "default_provider": "<name>"
}

default_size must be valid for the configured provider and model. Do not assume 1024x1024 is universally accepted. For the current volcengine configuration, use 2048x2048.

To add a new provider, just add an entry to providers. To switch default, change default_provider.

Running

python3 <skill-path>/scripts/generate.py \
  --prompt "<prompt>" \
  --output "<output_path>" \
  [--provider "<provider_name>"] \
  [--image "<input_image_path>"] \
  [--size "2048x2048"]

• --provider: optional, defaults to default_provider in config
• --image: enables img2img mode, size defaults to auto
• --size: override the provider's default size

Workflow

1. Run scripts/generate.py with prompt and output path
2. Pick a short descriptive filename (English, kebab-case)
3. Show the result to the user with the Read tool